fix: remove replica lag metric (#1696)

we can no longer measure replica lag because NPM's new protocol does not
allow us to query individual packages on the replica.

```
curl -H "npm-replication-opt-in: true" "https://replicate.npmjs.com/registry/construct-hub-probe"
"Not Found"%                                       
```

I can't think of a way to measure this now, so I propose to just rip it
out so that we can migrate to the new protocol

----

*By submitting this pull request, I confirm that my contribution is made
under the terms of the Apache-2.0 license*

---------

Signed-off-by: github-actions <[email protected]>
Co-authored-by: github-actions <[email protected]>

Author: kaizencc

src/__tests__/__snapshots__/construct-hub.test.ts.snap

@@ -589,20 +589,11 @@ export class NpmJs implements IPackageSource {
     const period = Duration.minutes(5);
 
     const alarm = new MathExpression({
-      // When the npm replica is sufficiently behind the primary, the package source will not be
-      // able to register new canary package versions within the SLA. In such cases, there is
-      // nothing that can be done except for waiting until the replica has finally caught up. We
-      // hence suppress the alarm if the replica lag is getting within 3 evaluation periods of the
-      // visibility SLA.
-      expression: `IF(FILL(mLag, REPEAT) < ${Math.max(
-        visibilitySla.toSeconds() - 3 * period.toSeconds(),
-        3 * period.toSeconds()
-      )}, MAX([mDwell, mTTC]))`,
+      expression: 'MAX([mDwell, mTTC])',
       period,
       usingMetrics: {
         mDwell: canary.metricDwellTime(),
         mTTC: canary.metricTimeToCatalog(),
-        mLag: canary.metricEstimatedNpmReplicaLag(),
       },
     }).createAlarm(canary, 'Alarm', {
       alarmName: `${canary.node.path}/SLA-Breached`,
@@ -708,24 +699,6 @@ export class NpmJs implements IPackageSource {
         ],
         rightYAxis: { min: 0 },
       }),
-      new GraphWidget({
-        height: 6,
-        width: 12,
-        title: 'Observed lag of replicate.npmjs.com',
-        left: [
-          canary.metricEstimatedNpmReplicaLag({
-            label: `Replica lag (${packageName})`,
-          }),
-        ],
-        leftAnnotations: [
-          {
-            color: '#ffa500',
-            label: visibilitySla.toHumanString(),
-            value: visibilitySla.toSeconds(),
-          },
-        ],
-        leftYAxis: { min: 0 },
-      }),
     ];
   }
 }

src/__tests__/devapp/__snapshots__/snapshot.test.ts.snap

@@ -29,14 +29,6 @@ export const enum MetricName {
    */
   TRACKED_VERSION_COUNT = 'TrackedVersionCount',
 
-  /**
-   * The estimated lag between the npm registry replica (replicate.npmjs.com)
-   * and the primary registry (registry.npmjs.com). This cannot be measured
-   * directly because the primary does not expose the relevant CouchDB endpoints,
-   * so we use the probe package to get a low-resolution view of this.
-   */
-  NPM_REPLICA_LAG = 'EstimatedNpmReplicaLag',
-
   /**
    * A metric tracking whether the npm registry replica (replicate.npmjs.com)
    * is down. The value is 1 when the replica is detected to be down, and 0

src/package-sources/npmjs.ts

@@ -85,26 +85,6 @@ export class NpmJsPackageCanary extends Construct {
     });
   }
 
-  /**
-   * The estimated lag between the npm registry replica (replcate.npmjs.com) and
-   * the primary registry (registry.npmjs.com).
-   *
-   * IMPORTANT NOTE: This is based on the difference in modified timestamps for
-   * the probe package between the two and hence has a granularly no better than
-   * the publishing interval of this. Since the construct-hub-probe package is
-   * only published every 3 hours approximately, this metric has a resolution
-   * that is strictly worse than 3 hours.
-   */
-  public metricEstimatedNpmReplicaLag(opts?: MetricOptions): Metric {
-    return new Metric({
-      period: Duration.minutes(5),
-      statistic: Statistic.MAXIMUM,
-      ...opts,
-      metricName: MetricName.NPM_REPLICA_LAG,
-      namespace: METRICS_NAMESPACE,
-    });
-  }
-
   /**
    * A metric tracking whether the npm registry replica (replicate.npmjs.com)
    * is down. The value is 1 when the replica is detected to be down, and 0

src/package-sources/npmjs/canary/constants.ts

@@ -25,7 +25,7 @@ const REPLICA_REQUEST_TIMEOUT_MS = 30_000;
 /**
  * This package canary monitors the availability of the versions of a specified
  * package in the ConstructHub catalog. It publishes metrics that help
- * understand how much time passes between a pakcage appearing in the public
+ * understand how much time passes between a package appearing in the public
  * registry and it's availability in the ConstructHub instance.
  *
  * From the moment a package has been published, and until it appeared in
@@ -74,8 +74,6 @@ export async function handler(event: unknown): Promise<void> {
     updateLatestIfNeeded(state, latest);
 
     try {
-      const replicaLag = await stateService.npmReplicaLagSeconds(packageName);
-
       await metricScope((metrics) => async () => {
         // Clear out default dimensions as we don't need those. See https://github.com/awslabs/aws-embedded-metrics-node/issues/73.
         metrics.setDimensions({});
@@ -89,16 +87,6 @@ export async function handler(event: unknown): Promise<void> {
           (await stateService.isNpmReplicaDown()) ? 1 : 0,
           Unit.None
         );
-
-        // If we weren't able to calculate the replica's lag, then simply
-        // don't report the metric.
-        if (replicaLag !== undefined) {
-          metrics.putMetric(
-            MetricName.NPM_REPLICA_LAG,
-            replicaLag,
-            Unit.Seconds
-          );
-        }
       })();
 
       for (const versionState of [
@@ -384,63 +372,6 @@ export class CanaryStateService {
     }
   }
 
-  /**
-   * Estimate how far behind the NPM replica is compared to the live NPM
-   * registry. If the NPM replica is down, return undefined.
-   */
-  public async npmReplicaLagSeconds(
-    packageName: string
-  ): Promise<number | undefined> {
-    const encodedPackageName = encodeURIComponent(packageName);
-
-    console.log(`Measuring NPM replica lag using ${packageName}...`);
-
-    const primaryDate = await getModifiedTimestamp(`registry.npmjs.org`);
-
-    let replicaDate;
-    try {
-      replicaDate = await getModifiedTimestamp(
-        `replicate.npmjs.com/registry`,
-        REPLICA_REQUEST_TIMEOUT_MS
-      );
-    } catch (e) {
-      if (e instanceof Error && e.message.includes('HTTP 504')) {
-        console.log(
-          `Warning: error fetching replicate.npmjs.com: ${e.toString()}`
-        );
-        // There is no value to report
-        return undefined;
-      } else {
-        throw e;
-      }
-    }
-
-    const deltaMs = primaryDate.getTime() - replicaDate.getTime();
-
-    console.log(`Timestamp on primary: ${primaryDate.toISOString()}`);
-    console.log(
-      `Timestamp on replica: ${replicaDate.toISOString()} (${
-        deltaMs / 3_600_000
-      } hours behind)`
-    );
-
-    // We return in seconds... The millisecond resolution is silly here since the probe package is
-    // only published approximately once every three hours. We use seconds only because this is the
-    // largest available time unit in CloudWatch.
-    return deltaMs / 1_000;
-
-    async function getModifiedTimestamp(
-      baseUrl: string,
-      timeoutMillis?: number
-    ) {
-      const isoDate = await getJSON(
-        `https://${baseUrl}/${encodedPackageName}`,
-        { jsonPath: ['time', 'modified'], timeoutMillis }
-      );
-      return new Date(isoDate);
-    }
-  }
-
   private key(packageName: string): string {
     return `${ObjectKey.STATE_PREFIX}${packageName}${ObjectKey.STATE_SUFFIX}`;
   }
@@ -534,6 +465,7 @@ function getJSON(
           headers: {
             Accept: 'application/json',
             'Accept-Encoding': 'identity',
+            'npm-replication-opt-in': 'true', // can be deleted after May 29: https://github.com/orgs/community/discussions/152515
           },
           timeout: timeoutMillis,
         },